Keaun Amani Podcast Transcript
Keaun Amani joins host Brian Thomas on The Digital Executive Podcast.
Welcome to Coruzant Technologies, Home of The Digital Executive podcast.
Brian Thomas: Welcome to The Digital Executive. Today’s guest is Keaun Amani. Keaun Amani stands at the forefront of blending the intricate worlds of software engineering and molecular biology as the CEO and founder of Neurosnap. With a profound passion that navigates the crossroads of these disciplines, Keaun leverages a unique amalgam of skills to spearhead innovative products that accelerate discovery in both realms.
His journey in the tech landscape is marked over nine years of intensive software engineering underpinned by a vast coding expertise and demonstrated a prowess in integrating machine learning models into practical applications. Beyond his technical acumen, Keaun is celebrated as a polymath in biology, with a command over molecular, plant, synthetic, and computational biology.
This diverse experience makes him one of the rare scientists who seamlessly transition between academia and industry, pushing the boundaries of knowledge and application in both biology and software.
Well, good afternoon, Keaun. Welcome to the show.
Keaun Amani: Thank you very much for having me.
Brian Thomas: You better believe it. It’s an awesome day to start a podcast. I appreciate you making the time Keaun. And Keaun, jumping into your first question here, your work uniquely combines software engineering with molecular biology. How did you develop this interdisciplinary approach and what inspired you to bridge these two fields?
Keaun Amani: This is a fantastic question, and honestly you know, going back to when I was much younger, I’ve always found molecular biology, as well as you know, programming, computer science in general, very interesting. They’ve always been, like, a long-time passion of mine. You know, growing up, I did everything I could to really learn more and more about these subjects.
And eventually, when I got to university, I decided to go about doing a project where we would genetically modify, like, some plants to actually express some genes that would make them bioluminescent. So, this was a really interesting project because you know, plants, while a lot of us think of them as kind of like these things you do for gardening or kind of like as like a part time hobby, For molecular biologists, they’re truly, truly remarkable organisms.
They can be used to produce all sorts of different compounds and chemicals. There are some groups that are even working on genetically engineering plants to like to produce certain types of vaccines or certain types of like peptides and therapeutics. And on top of that, you know understanding plant molecular biology is critical to a lot of the problems that we’re facing in like agriculture and stuff like that.
This was really an interesting subject for me. And the more I got into plant molecular biology, the more I realized a lot of this could actually be automated. In fact, a lot of the research that I was doing could have been streamlined using a lot of these different machine learning models and tools. You know, this, this really fueled my passion for these subjects.
And I essentially decided to combine my knowledge of both fields and try to bridge them as much as possible by applying my knowledge of data science and computer science to the large data sets that we had. For protein engineering and as well as for like just genetically engineering organisms.
Brian Thomas: That’s awesome. And, you know, we talk a lot about some of the innovations and emerging tech science here on the podcast, but I really do see this as a unique podcast today, combining these two fields. And I appreciate you breaking that down for our audience. Keon, Neurofold has significantly outperformed existing benchmarks in enzyme design models.
Could you explain the key innovations behind Neurofold and its impact on the scientific community and industry?
Keaun Amani: Yeah, absolutely. So just to give a little bit of background for our audience. Neurofold is an enzyme design model, basically helps people optimize certain types of enzymes. And the reason why this is important is because in industry, enzymes are extremely valuable.
They tend to facilitate and catalyze certain types of reactions. So, let’s say you wanted to create something like, like a bioreactor, or you wanted to mass produce certain types of compounds. Or let’s say you have a detergent, and you just want enzymes that work in like elevated pH environments and elevated, you know, temperatures that you’d have in like a washing machine.
Then this is where like enzyme design really comes into place. The only problem with this is that in order to do an enzyme optimization, kind of like project that typically takes years’ worth of time as well as cost millions and millions of dollars, and you’re not really guaranteed to get any results.
So, finding a way to reliably optimize enzymes with elevated thermostability and pH stability as well as like reaction rate, this is really crucial in industry. And this is exactly what we aim to solve with our Neurofold model. In terms of actually answering the question now, the major innovation behind Neurofold is mostly due to the fact that we adopt a multimodal approach that was, that was not really done in this field before.
So, for those who don’t know, multimodal models are similar to models like DALI and other image generation models where they accept more than one type of modality. A modality is just like a type of data or a type of information. So, in terms of like DALI, the two types of modalities that they accept. They took in originally were text data as well as image data.
So, when you combine those two sources of information, the model that you’re training not only learns a better representation of what reality looks like, but also learns more about each one of those different modalities as well as how to combine them. And it makes the model like a lot more powerful in terms of use cases.
So, it’s basically the same thing with biology. We’re in the case of neuro fold, the modalities that we combine, we combine information from the protein structure, as well as evolutionary information of the protein, as well as just a single sequence approach. And we did this in a very interesting way, which basically allows our model to infer about like the protein fitness landscape with a much greater degree of accuracy than what was previously achieved.
So, for example, I know Meta slash Facebook, they produced a model called the ESM One V a couple of years ago. And when benchmarked against that model in terms of Spearman rank correlation, our model is 40 times better. If I remember correctly, that was one really, really exciting accomplishment that came out of like the Neurofold paper. And we’re really excited about that.
Brian Thomas: That’s amazing. And I appreciate you sharing that, Keaun. That’s just awesome. And the fact that, you know, being benchmarked across some of these large companies that do a lot of this work and have a lot of money behind them. It’s just simply amazing that you’re able to do this.
And I appreciate you breaking that down really for our audience today. So, thank you. And Keaun, you’ve successfully integrated machine learning models into practical applications in molecular biology. How do these models enhance research and development? And what future advancements do you foresee in this integration?
Keaun Amani: That’s another really great question. So personally, If I were to make some ambitious predictions, I would say that these models can be applied to virtually any kind of research area or subdiscipline within the molecular biology field. And the reason why I say this is because you know, you take something like protein designer, protein folding, or even just predicting the toxicity of certain types of substances.
This can be widely applied to a number of different things from therapeutic development to agriculture to even certain things in material science. So, I think that in terms of the number of different areas that this could affect can be very broadly applicable and in terms of future advancements. What I foresee is I foresee this to expand into other areas as well.
That might not even be super related to molecular biology. So, for example, different areas of chemistry could certainly benefit greatly from a lot of these different approaches as some of our tools and services are very focused on like individual small molecules and chemicals instead of even like you know, purely biological models like proteins and nucleotides.
Additionally, you know, there’s, there’s certainly a lot of applications as well for material science. Right now, we are already starting to actually expand our offering for material science in the form of like molecular dynamics, but as our platform grows and grows more, then we’ll certainly be able to branch out, you know, start targeting these different areas a little bit better as well.
Brian Thomas: Thank you. I appreciate that. You know, we’re infusing a lot of this AI and machine learning into our everyday lives, and we’re really advancing mankind and the services, the innovations, the platforms that we leverage every single day in our businesses through the course of machine learning and artificial intelligence.
So, I really do appreciate that. And Ken, last question of the day. Neurosnap is at the forefront of accelerating discovery and software engineering and molecular biology. What are your future goals for the company and how do you plan to continue pushing the boundaries of innovation?
Keaun Amani: That’s a good question.
I think antibodies are going to be playing a really big role in the future. In fact, even right now, they arguably are some of the best drugs and therapeutics that we have are antibody-based medications. And one of the reasons why they’re so great is because They tend to be much safer in terms of side effects, and on top of that, they do a really good job of working alongside your immune system, whereas like you know, some other substances, they’re, they’re more so designed to just inhibit something specific, you know, obviously there are, there are exceptions to this, and it gets a lot more complicated in real life, you know, antibody therapeutics are really, really cool.
The only problem is they’re also some of the hardest and most expensive types of drugs to develop. As It’s very complicated stuff and a lot of, you know, just, just designing like the antibody itself to make it like target something ambiguous. This is nontrivial, and this could be very resource intensive, resource and time intensive.
What we’re going to do, and what we have been doing for the last several months, Is we’re going to be kind of doubling down on services centered around antibody design. We’re really going to be focusing on developing different tools and models that are going to be accelerating different points within like the research and development pipeline for the development of these antibody therapeutics.
I think in the long run, this is going to be extremely beneficial, not just to the end users, like the end recipients of these medications and therapeutics, but on top of that, I think it’s going to be very beneficial to the industry. As it’ll help build trust in biotech and pharmaceutical industry.
And on top of that, another benefit is I think it will be really good in terms of bringing more of us and like investor attention into these fields as well.
Brian Thomas: Thank you. Do appreciate that. And I know you have a lot of experience in this field, breaking that down, you know, I, I was a software engineer early in my career, but doing the biology stuff is, is not my forte, but I appreciate you breaking that down with the antibody therapeutics.
Keaun, just want to let you know, it was such a pleasure having you on today and I look forward to speaking with you real soon.
Keaun Amani: Yeah, thank you very much. It was an absolute pleasure to answer these questions. These were all really fantastic questions. Thank you for having me. I look forward to the next time we talk.
Brian Thomas: Bye for now.
Keaun Amani Podcast Transcript. Listen to the audio on the guest’s podcast page.
